The most popular AI coding tools right now

Ananya Kapoor, software engineer at financial services company Stripe, uses Claude Code every day.

Sometimes she’s leveraging the AI assistant to explore a new idea; other mornings, she’s relying on it to investigate an error. The tool helps her understand how different components of code interact, and how those connections may lead to unexpected results.

“It’s pretty helpful…in terms of going in and really analyzing the code base,” Kapoor said, “because I think that is what is really tough for humans.”

And Kapoor is not alone in her choice of AI tool. According to an annual survey from software engineering intelligence platform Jellyfish, Claude Code is the most popular AI coding tool; Almost four in 10 (39%) of 636 surveyed global engineering professionals said they use it. Other top choices included Gemini Code Assist (35%), GitHub Copilot (31%), and OpenAI Codex (20%).

Top reasons for using these tools, according to the poll, included code writing, code review, and code explanation. We spoke with other users of the leading tools to learn more about what they’re generating at work.

Claude Code (Anthropic) allows users to find dependencies in unfamiliar code, create and edit files across codebases, execute commands, and run tests. A recent update includes a desktop version to manage parallel sessions.

Kapoor gives Claude persistent instructions through a set of guidelines known as a CLAUDE.md file. One essential directive in her specific .md file includes a command to the effect of, “Don’t make any changes without asking me.”

“It’s better if it’s just one place where you have all those instructions,” Kapoor said. Kapoor added that she has an account for work as well as a personal subscription to perform tasks related to fitness tracking, email proofreading, and general ideation.

Gemini Code Assist (Google) supports features like code completion and generation, a conversational assistant, and agentic chat to complete multi-step tasks. A recent update included Outlines, a generation of English summaries integrated with source code, and a pair-programmer-like assistant called Finish Changes.

Gemini allows users to provide images as input. Lokesh Karanam, staff software engineer at marketing and business services RRD, uses Gemini Code Assist (or specifically the command-line interface option known as Gemini CLI) for “web-heavy” interfaces and websites.

“Let’s say, if I’m working on a simple feature like adding a button or something, I use Gemini CLI, because Gemini can read images,” Karanam said, noting the tool sometimes lacks speed in providing outputs compared to other tools.

GitHub Copilot announced itself as an “AI pair programmer” in June 2021.

Karanam uses GitHub Copilot (again the CLI version) for work purposes. He frequently employs the tool’s “/plan” option, which analyzes a request to create an implementation checklist. Lately, Karanam has also been telling Copilot: “Don’t give me what I think I need. Give me what I actually need.” (On top of that, he asks the tool to ask clarifying questions during the coding process.)

“Ninety-five percent of the time it gives me what I need. I just need to make some simple changes,” he said.

He appreciates that Copilot allows him to use a variety of models, including recent updates to ChatGPT, Claude, and Gemini. A simple refactoring or changing of a variable, for example, may not require a sophisticated reasoning model—an important consideration as providers consider new pricing options based on token usage.

Codex (OpenAI) arrived in May 2025, with features like proposing pull requests for review, answering codebase questions, and fixing bugs. (OpenAI announced a mobile version just last week.)

Already a longtime ChatGPT user, Andrew Southall, founding engineer at cybersecurity company SkySiege, appreciates the Codex capabilities available in a basic OpenAI subscription—and what they can do for his company’s website.

Southall runs Codex on isolated servers to support the company’s front-end and website work. “I just need it to do the things that are going to save me hours of sitting around typing basically,” he said.

Change request. Some developers like Karanam don’t see too much of a difference between code assistant tools like GitHub Copilot and Gemini Code Assist. Will Holtz, COO and CFO of marketing insights platform Prescient AI, said he has Claude and Codex and he’s “almost agnostic to some of the tools at this point.” Some engineers like Kapoor, however, appreciate the familiarity of one go-to tool.

“If it’s not substantially better, or if it’s not solving a problem that currently this tool isn’t solving for me, why would I want to switch?” Kapoor said.